Background The gut-liver axis plays a critical role in liver disease progression; however, how gut microbial ecology and function vary across disease stages remains unclear.

Objective To define stage-specific microbial and functional signatures and evaluate their diagnostic potential.

Design We analysed faecal samples from 1168 individuals spanning healthy controls, fatty liver, hepatitis, cirrhosis and hepatocellular carcinoma by 16S rRNA sequencing, with a subset (n=141) profiled by shotgun metagenomics. To increase statistical power and enable external validation, 2376 publicly available metagenomic datasets, including 734 liver-related, were integrated. Machine learning-based multicohort analysis was used to identify microbial biomarkers, assess risk factors and classify disease stages.

Results Microbial diversity declined and a low-richness enterotype expanded with disease severity. Machine learning revealed a discordance in hepatitis, which lacked taxonomic markers but was defined by a conserved functional signature of biosynthetic upregulation. In contrast, advanced stages featured consistent markers like Ligilactobacillus and Veillonella, with strain-level evidence confirming oral-gut transmission. Functional profiling delineated a metabolic continuum from anabolic precursor synthesis in hepatitis to virulence factor production in cirrhosis and putrefactive metabolism in carcinoma. Comparative analysis confirmed that these signatures were distinct from those in non-liver metabolic and oncologic disorders. Importantly, the expansion of oral-derived Veillonella spp and the low-richness enterotype were significantly associated with increased mortality.
Conclusion This large-scale study delineates stage-dependent ecological and functional remodelling of the gut microbiome across liver diseases. These findings highlight the potential of microbiome-based markers for non-invasive diagnosis and prognostic risk stratification in liver diseases.